When semantic judgment becomes cheap enough to put everywhere (sameernanda.com)

TypeSafe AI has introduced Jev, a new semantic judgment model released in early access, that offers a cost-effective alternative for obtaining structured answers from text inputs. Unlike traditional language models like ChatGPT, which generate conversational text, Jev functions similarly to a multiple-choice exam, returning probabilities for various options rather than generating prose. Its unique primitive, known as “Noul,” provides a probability of “yes” rather than a simple true/false output, empowering developers to make informed decisions based on specific thresholds for action. The significance of Jev lies in its affordability and quick response times, costing only $0.042 per million input tokens with no output charges and boasting response times of 70–500 milliseconds. This efficiency allows for broader application in scenarios like job search pipelines, where nuanced language can complicate decision-making. By integrating Jev into workflows, developers can leverage its semantic judgment to handle ambiguous statements and complex queries efficiently, promising to streamline tasks that were traditionally challenging and time-consuming for more generic models. As semantic judgment becomes quicker and cheaper, it has the potential to become a fundamental component in software development, encouraging its widespread adoption across various applications.